ricks

ricks is an OpenCode-native consensus plugin that lets you invoke @ricks to run a multi-model review loop on a single task.

Intro

ricks is inspired by the Council of Ricks from Rick and Morty: multiple independent "Ricks" review the same problem, challenge weak spots, and keep iterating until they align.

In this plugin, that idea becomes a practical harness for OpenCode:

• The primary model creates the initial answer.
• Selected reviewer models evaluate it one-by-one.
• Each reviewer must return either YES (approve as-is) or a fully revised plan.
• Revisions are passed forward to the next reviewer.
• The process repeats across rounds until all selected reviewers return YES in the same pass (or round limit is hit).

This project is intentionally OpenCode-only: model discovery, provider access, and auth are delegated to OpenCode.

Key capabilities

• Invoke with @ricks in normal chat flow.
• Open setup with @ricks setup.
• Model toggle list with inline reasoning override per model.
• Select all / deselect all flows.
• Persistent project settings in .opencode/ricks.settings.json.
• Continuous observable progress with a live event timeline (agree/disagree/error, per model, per round).
• Final stats summary (approvals/revisions/errors + per-model breakdown).
• Per-round outputs in the final report.
• Setup markdown prefill so users can edit and submit config in one message.

How consensus works

1. Initial plan generation

• Primary model (build agent) produces a full initial plan for the user task.

2. Sequential reviewer pass

• For each selected model in order, ricks asks:
• approve with exactly YES, or
• provide REASON + full revised PLAN.

3. Plan handoff

• If reviewer says YES, plan stays unchanged.
• If reviewer revises, that revised plan becomes the current plan for the next reviewer.

4. Round completion

• If all reviewers returned YES in that round, consensus is reached.
• Otherwise, next round starts using latest revised plan.

5. Final output

• Returns final plan plus summary, stats, timeline, and per-round details.

Install

ricks is published as an npm package: opencode-ricks.

Add it to your OpenCode config:

{
  "plugin": ["opencode-ricks"]
}

Optional: pin a version explicitly:

{
  "plugin": ["[email protected]"]
}

Usage

Run consensus on a task:

Ask @ricks for a rollout plan for migrating our monolith to services.

Open setup:

@ricks setup

Setup shortcuts

@ricks setup select all
@ricks setup deselect all
@ricks setup toggle openai/gpt-5
@ricks setup reasoning openai/gpt-5 high
@ricks setup max rounds 5
@ricks setup strategy normalized
@ricks setup timeout 120000
@ricks setup hide by default on
@ricks setup save as default on

One-shot setup via markdown template

@ricks setup can prefill your input with editable markdown (models/options). Submit it directly to apply all changes at once.

Settings

Stored per project in:

.opencode/ricks.settings.json

Fields include:

• selectedModels
• modelVariants (reasoning overrides)
• maxRetries (internal storage; corresponds to user-facing max rounds - 1)
• strategy (exact | normalized | json)
• timeoutMs
• saveAsDefault
• hideByDefault
• initialized

Runtime config (optional)

You can define defaults in opencode.jsonc:

{
  "ricks": {
    "models": ["openai/gpt-5", "anthropic/claude-sonnet-4-5"],
    "strategy": "normalized",
    "maxRounds": 100,
    "timeoutMs": 120000
  }
}

Notes:

• Current default maxRounds is 100.
• Setup-managed settings generally drive normal @ricks runs.
• Only connected/available models are presented in setup and used for execution.

Output shape

A consensus result includes:

• Final answer
• Summary (task, rounds, consensus result)
• Ricks stats
• Timeline (continuous model events)
• Per-round outputs (YES/revision/error by model)

Development

npm install
npm run typecheck
npm run build

For local development/testing, load the built plugin directly:

{
  "plugin": ["file:///absolute/path/to/ricks/dist/index.js"]
}

License

MIT